Friday morning I asked my AI to publish a new paid story on Substack. It couldn't. Instead of faking a post, it found the real blocker, fixed the engine that feeds it source material, and published the real thing. That honest recovery is the whole story.
I give my AI one job every Friday morning.
Publish a new paid post.
Not a note. A full article. The kind you get when you pay for Liz on the Web. Fresh story. Fresh prompt. Fresh proof that the thing works.
I hand it one instruction and walk away.
The whole machine is supposed to just run.
Last Friday at 8:04 AM, it didn't.
My AI tried to write a story and hit a wall it had never hit before.
The source queue was empty.
Every single source packet I had written was already published. There was nothing left to turn into a new story. No links I had saved. No research I had clipped. No "here is a real thing I built" material sitting in the queue.
That sounds like a failure.
It was not.
Because instead of inventing a fake win and pretending the post went live, my AI did something smarter. It showed me the blocker. It traced back through the posting watchdog that catches broken lanes. It found the permission issue that was preventing the refill script from adding new source packets. It fixed that issue. It added a real sourced packet to the queue. It ran the author and publisher the right way. And then the post went live.
That is the whole recovery.
The weird part is that this honest failure is more useful than a fake success would have been.
Because now I know that when Friday posts go sideways, the system will tell me why. It will not make something up and charge my subscribers for it. It will find the real problem and hand it back to me.
That is the kind of AI work that matters.
What this article gives you
You get the full breakdown of what happened when the Friday post broke. You get to see what the watchdog caught, where the permission block lived, how the refill worked, why the queue was empty, and how a new source packet got added.
You also get the prompt I use to kick off the whole Friday machine, what each step checks for, where my AI needs me to be ready with a decision, and a step-by-step for how you can set up your own honest failure recovery.
Who this is for
This is for you if:
You publish on a schedule and want to know when that schedule breaks.
You use AI to run a repeating task and want proof it actually ran.
You automate content or business work and need to know when the automation hits the edge.
You want your AI to fail honest instead of fail pretty.
Why you can trust this story
The watchdog sent Mission Control task 8775 at 8:04 AM. The task said the author lane had no source material. That same day, the macmini recovery script ran as task 8778 and measured the disk, checked the file attributes, and repaired the scoped permissions the author wrapper needed. The author tried again and logged "queue-empty" at 8:17 AM. That is not a success. That is the real blocker.
The publisher searched for a staged For-Me draft newer than Part 48 and found nothing, because there was nothing to publish yet. The fix was a new sourced packet added to the queue. Then the author ran a second time. Then the publisher ran. Then the post went live.
No fake story. No invented receipts. Just the machine working honest and me being able to see where it broke.
Behind this paywall
Paid readers get:
The full watchdog alert from when the lane failed and what it told me to check first.
The permission block that stopped the refill script from running.
The exact moment the queue measured empty and where that check lives in the code.
How the new source packet got added and why it mattered.
The eight-step flow from "post broke" to "post lives" without faking any step.
What I had to do by hand and what the machine did by itself.
The one mistake that could have caused the publisher to go silent on retry.
A copyable prompt for setting up your own broken-lane detection and fix.
What paid subscribers get every week
Paid subscribers do not only get this one posting recovery. You also get:
Real work I hand to my AI inside my actual business.
The exact prompts I use, written so you can copy them.
The receipts that prove what worked.
The problems that slowed the job down.
The decisions I had to make that the AI could not.
Simple playbooks you can use in your own work.
Subscribe now and the rest of this article opens right away. You get the full recovery story, the prompt, the checking steps, and the broken-lane checklist now.
The full walkthrough is below the paywall.
Click the button, subscribe, and the locked section below opens right away. You get this article now.
